February 2026 Summaries
4 posts from Polar Signals
Filter
Month:
Year:
Post Summaries
Back to Blog
Polar Signals Cloud initially had a straightforward authentication and authorization model, where human users could either perform all actions or just query data, and machines were limited to specific project-based operations. However, this setup fell short for enterprise customers with diverse requirements, such as managing billing or accessing only certain projects. To address this, Polar Signals developed a more flexible identity and access management (IAM) system, where every action has distinct permissions bundled into roles that can be associated with identities, either organization-wide or project-specific. Humans authenticate via OIDC, while machines use service accounts. The new system includes predefined roles reflecting common use cases, and all previous permissions have been migrated, with new role creation under the old system disabled. Migration guides are available for different use scenarios, and while there is no end-of-life date for the old project tokens, users are encouraged to upgrade. The development benefited from contributions by Turbopuffer and advice from Eric Chiang of Oblique.
Feb 24, 2026
461 words in the original blog post.
Understanding the call stack is crucial for programming, as it manages function calls and returns in a computer's memory, allowing for the continuation of code execution. This text delves into how native-compiled code, specifically on 64-bit ARM Linux systems, manages call stacks, emphasizing the role of stack frames and the use of registers like x29 and x30 for managing function call returns. Tools such as profilers and debuggers rely on accurately interpreting these call stacks to perform tasks like debugging and profiling, often using techniques like frame pointers or the .eh_frame section for stack unwinding. While frame pointers offer a straightforward way to trace call stacks, they are often omitted to optimize performance, necessitating alternative methods like .eh_frame for accurate stack tracing. The document concludes by hinting at further exploration into how profilers are triggered and handle virtual-machine environments, promising more insights in future discussions.
Feb 19, 2026
2,985 words in the original blog post.
Since the launch of the MCP, users have faced challenges accessing it due to the requirement of creating a project token, which necessitates an "Owner" role, limiting access primarily to privileged users. This token system was initially intended for data pushing to Polar Signals Cloud but was misappropriated for MCP, creating a paradox where users could query data but needed higher privileges to use MCP. To address this, MCP has adopted Oauth, which allows users to authenticate using their identity to issue profiling data queries, eliminating the need for long-lived tokens and manual invalidation. This implementation simplifies access, enabling users to query profiling data through their preferred MCP clients using Oauth, with guidance provided through an in-product documentation page for various AI integrations, such as Claude Code and Cursor. Further authentication and authorization improvements are anticipated.
Feb 12, 2026
411 words in the original blog post.
At FOSDEM, the team delivered four presentations covering topics on database evolution, Rust application profiling, deterministic simulation testing, and Go heap memory profiling, showcasing significant advancements in these areas. They discussed transitioning their database to a cloud-native model using technologies like Vortex and Delta Lake, which reduced costs by 50% and improved query latency by 200%. The profiling of Rust applications was demonstrated through Parca, an open-source continuous profiler that utilizes eBPF to minimize overhead, with features such as jemalloc integration. Deterministic Simulation Testing (DST) in Rust was introduced to ensure distributed system reliability by using state machines and random seeds for simulating failures. Additionally, OOMProf, a Go library, was presented as a tool for capturing a memory profile at the time of an out-of-memory event to aid in debugging, thanks to its use of eBPF to intercept kernel signals. The event was a platform for exchanging innovative ideas and connecting with the community, with an open invitation for further discussions via Discord.
Feb 10, 2026
510 words in the original blog post.